Pith. sign in

REVIEW 3 major objections 4 minor 129 references

This paper proves that a neural-operator extension of Neural Jump ODEs converges to the conditional expectation—the L2-optimal predictor—for stochastic processes taking values in a function space.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 03:34 UTC pith:AQF6AMMC

load-bearing objection Solid infinite-dimensional extension of NJ-ODE, but the advertised "measurability" weakening is really an absolute-continuity assumption, and the experiments don't exercise the between-observation dynamics. the 3 major comments →

arxiv 2607.23110 v1 pith:AQF6AMMC submitted 2026-07-25 stat.ML cs.LG

Operator Neural Jump ODEs: L²-optimal prediction in function spaces

Florian Krach , Oliver L\"othgren , Josef Teichmann This is my paper
classification stat.ML cs.LG
keywords neural jump ODEsoperator learningconditional expectationfunction-valued stochastic processesL2-optimal predictiongeneralized kernelconvergence guaranteesonline learning
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper extends Neural Jump ODEs to predict stochastic processes that take values in an infinite-dimensional function space, rather than a finite-dimensional vector space. The proposed Operator NJ-ODE uses a generalized kernel to aggregate observations at random spatial locations and a recurrent hidden state to store past information. The main result proves that, as the model complexity grows, the trained model converges to the conditional expectation of the process—the L2-optimal predictor—with respect to a metric that accounts for the random observation scheme. The paper also proves that the Monte Carlo approximation of the loss converges, and it weakens previous continuity assumptions to measurability together with an integrability condition.

Core claim

The paper claims that the Operator NJ-ODE is the first member of the NJ-ODE family with convergence guarantees for function-valued processes in L^2(Ξ,R^{dX}), without finite-dimensional discretization. The conditional expectation is shown to be the unique minimizer of a natural loss function up to indistinguishability, and the model, with growing complexity, approximates this minimizer arbitrarily well in the training loss and in the metric d_k. This is achieved by an approximation argument that uses the universal approximation theorem on carefully chosen finite-dimensional subspaces, while letting the truncation dimensions grow with the network complexity.

What carries the argument

The central object is the generalized kernel ψθ3, which aggregates an arbitrary number of spatial observations at each observation time into a fixed-dimensional input, and the recurrent hidden state that carries the history of past observations. The proof relies on representing the conditional expectation through a measurable function F and its generalized derivative f, then approximating both by neural networks on ε-bounded subspaces where the number of observations is truncated. This yields a constructive architecture that provably approximates the optimal predictor.

Load-bearing premise

The load-bearing assumption is that the conditional expectation process has a generalized derivative f such that it can be written as an integral of f between observation times—an absolute-continuity-type condition that is not automatically satisfied by measurable processes.

What would settle it

A controlled experiment where the target process has a conditional expectation that jumps at observation times (so no integral representation holds between observations), while all other assumptions are satisfied, and observing whether the O-NJ-ODE still converges to the conditional predictor as network complexity grows.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the theorem holds, function-valued processes such as yield curves, volatility surfaces, or EEG signals can be predicted online with an L2-optimal guarantee, without discretizing the output space.
  • The convergence in the metric d_k means the model learns the conditional expectation at all observable times and locations, including left-limit values at jumps.
  • The Monte Carlo convergence result justifies training on a finite dataset of irregular, incomplete observations.
  • The framework generalizes previous NJ-ODE results to infinite dimensions and weakening the assumptions to measurability expands the class of target processes covered, provided Assumption 3 holds.
  • Extensions to noisy observations, dependent observation frameworks, and input-output systems transfer to this setting as noted in the paper.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A natural extension is to replace the pointwise evaluation approach with a single hidden state that encodes the whole function, trading computational cost for a more efficient inference procedure.
  • If Assumption 3 fails—for example, for processes whose conditional expectation is not absolutely continuous between observations—the convergence guarantee would not apply, so checking this condition for real-world data becomes important.
  • The proof technique of ε-bounded truncation could be adapted to other infinite-dimensional learning problems where universal approximation is needed on unbounded input spaces.
  • The paper's claim of measurability being sufficient is only true in conjunction with Assumption 3; without it, the representation of the conditional expectation through a generalized derivative is not guaranteed.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper extends the Neural Jump ODE (NJ-ODE) framework to stochastic processes taking values in H = L^2(Ξ, R^{d_X}). The proposed Operator NJ-ODE (O-NJ-ODE) is a recurrent model with a neural-ODE evolution between observation times and a generalized integral-kernel jump update at observation times. The main theoretical result, Theorem 4.1, states that under Assumptions 1–7 the minimal training loss converges to the minimal value of the population objective, which is uniquely attained (up to indistinguishability) by the conditional expectation process, and that the trained model outputs converge to the conditional expectation in the pseudo-metrics d_k. Theorem 4.4 gives almost-sure uniform convergence of the Monte Carlo loss and a data-dependent scheme that achieves d_k-convergence. The proof relies on an L^p universal approximation argument for the conditional-expectation function F and its assumed generalized derivative f, combined with truncation of the unbounded observation input.

Significance. If the main theorem were valid under the stated assumptions, the paper would be a substantial step forward: it provides the first NJ-ODE-type convergence guarantee for function-valued outputs without finite-dimensional discretization, and it includes a complete proof structure, a public codebase, and synthetic experiments on a Brownian cosine field. The authors are careful to state their architecture and training objective. However, the central advertised message that 'measurability is sufficient' is not supported by the paper's own Assumption 3, which is an absolute-continuity-type integral representation. This narrows the actual contribution: the theorem applies to conditional-expectation processes that are absolutely continuous between observation times, not to general measurable (e.g., càdlàg) conditional expectations. With an honest revision of the claims, the paper still offers a valuable convergence theorem for a meaningful class of function-valued processes, but the current overstatement is load-bearing and needs to be corrected.

major comments (3)
  1. [Section 2.2, Assumption 3; Abstract; Introduction; Conclusion] The paper repeatedly claims that the new proof only needs measurability of the conditional-expectation function, and Remark 2.2 says continuity is weakened to integrability. But Assumption 3 requires, for fixed past observations, F(t,O[0,τ(t)]) = F(τ(t),O[0,τ(t)]) + ∫_{τ(t)}^t f(s,O[0,τ(t)]) ds P-a.s. in H. This is an absolute-continuity condition on t ↦ F(t,O[0,τ(t)]) between observation times; it is not implied by Doob–Dynkin measurability. A deterministic càdlàg process X_t(ξ)=1_{t≥a} φ(ξ) with observation times independent of X and having continuous distribution satisfies Assumptions 1, 2, 4, 5, 6, 7, but the conditional expectation jumps at a, so no integrable f can represent it. Hence Theorem 4.1 does not cover simple jump/regime-switch processes, despite the advertised generalization. The conclusion (Section 6) itself lists discontinuous processes as future work, which is inconsis
  2. [Theorem 4.1, Step 1, Eq. (14)] In the chain proving uniqueness, the display replaces the average over spatial points, (1/J_k)Σ_{j=1}^{J_k}, by the single j=1 term and keeps the inequality sign. This is not valid as a pointwise inequality: an average can be smaller than its first term. The subsequent text invokes exchangeability of the indices conditional on J_k to justify eliminating the J_k factor. If the intended argument is an equality after taking conditional expectations, this should be stated and proved before the display. As written, the uniqueness proof has a gap at exactly the step that converts the quadratic deviation into the metric d_k. This is fixable, but it is load-bearing for the conclusion that the minimizer is unique up to indistinguishability.
  3. [Definition 3.1/eq. (4) and Theorem 4.1, Step 2] The model definition and Algorithm 1 use self-imputed observations X̃_j^{t_k} = M_j^k⊙X_{t_k,ξ_j^k} + (1−M_j^k)⊙Y_{t_k-}(ξ_j^k) as inputs. However, the proof in Step 2 treats the generalized kernel's input as the truncated raw observation collection O^ε_{κ(t)}: 'ψ̃θ3(t,ξ,O^ε_{κ(t)}) is simply a projection on the truncated input'. Section 3.2 likewise defines the sampled O^{(l)}_{i,j} using the raw masked observation M_i^j⊙X, not the self-imputed value. The universal approximation argument approximates F as a function of the raw observations, so it does not directly apply to the recurrent self-imputed inputs used by the model. Either restrict the theoretically analyzed model to raw inputs, or add a step showing that the network can recover M⊙X̃ = M⊙X (e.g., by multiplying by the mask) so that self-imputation does not change the input distribution relevant to the UAT argument.
minor comments (4)
  1. [Throughout] The paper refers to 'Theorem 4.2' in Section 2.1 and in the proof of Theorem 4.1, but no Theorem 4.2 is stated; the intended reference appears to be Lemma 4.2. Please renumber or add the missing statement.
  2. [Section 4.1] The text first says Φ(θ) is lower semicontinuous via Fatou's lemma, then later says Φ is continuous on the compact set Θ_m. Since the solution of the ODE depends continuously on parameters, continuity is the stronger and sufficient statement; the Fatou justification is unnecessary and potentially confusing.
  3. [Remark 2.2] The wording 'weakened to integrability' is vague. Assumption 3 is not merely L^1-integrability of f; it is the existence of an integral representation. Please state precisely that F must be absolutely continuous between observation times with an integrable generalized derivative.
  4. [Section 5] The synthetic experiments only consider the continuous martingale process X_t(ξ)=W_t cos(ξ). Given the theoretical restriction just identified, an experiment with a jump or regime-switch between observation times would help delineate the actual scope of the convergence theorem.

Circularity Check

0 steps flagged

No significant circularity: the convergence proof applies UAT to functions defined independently of the model, and the loss gap controls the metric; self-citations are not load-bearing.

full rationale

The central chain is self-contained. The paper defines the target X_hat = E[X_t|A_t] via Doob-Dynkin as F(t,O[0,tau(t)]), then postulates in Assumption 3 an integral representation F(t,O)=F(tau(t),O)+∫ f(s,O)ds. The proof of Theorem 4.1 does not rename this representation as a prediction: Step 1 establishes that X_hat uniquely minimizes Ψ using the variance decomposition (Lemma 4.2) and a lower bound relating Ψ(η)-Ψ(X_hat) to d_k(X_hat,η). Step 2 approximates the pre-existing functions F and f by neural networks via Lp UAT (Hornik 1991) on growing truncated input spaces; the O-NJ-ODE is shown to approximate the composition through error bounds (21)-(23), giving Φ(θ*_m)→Ψ(X_hat), and (26) converts the loss gap into d_k convergence. No fitted parameter is renamed as a prediction: the loss minimizer's convergence is derived from the variance decomposition, and the UAT targets F and f are independent of model outputs. Citations to the authors' prior work (Krach et al. 2022 for the framework, Heiss et al. 2025 for Monte Carlo lemmas) support secondary or framework aspects; the main theorem's approximation argument is proven in the text. A genuine scope concern exists but is not circularity: Assumption 3 is an absolute-continuity/integrability condition, and the abstract/introduction claim that 'measurability is sufficient' overstates what is proved. Jump-type conditional expectations that fail Assumption 3 are outside the theorem, but that is an assumption/coverage gap, not an equivalence-by-construction. Hence score 1: no substantial circularity, with minor reliance on the authors' prior work.

Axiom & Free-Parameter Ledger

0 free parameters · 11 axioms · 0 invented entities

The theoretical claim rests mainly on the standard probabilistic and approximation-theoretic toolkit together with the seven assumptions. No new physical entities are postulated; the generalized kernel is an architectural component, not an invented entity with an independent falsifiable handle. The most fragile entry is Assumption 3, which postulates an integral representation for the conditional expectation process—this is an ad hoc regularity condition introduced to make the proof work, not a consequence of measurability.

axioms (11)
  • standard math Doob-Dynkin lemma guarantees a measurable function F with F(t,O[0,t]) = E[X_t|A_t]
    Invoked in Section 2.1 to justify the existence of the target function F.
  • standard math Lp universal approximation theorem for feedforward networks with bounded non-constant activation functions (Hornik 1991)
    Used in Step 2 of the proof of Theorem 4.1 to approximate F and f on ε-bounded subspaces.
  • standard math Existence and uniqueness of solutions to the O-NJ-ODE SDE (Protter 2005, Chap. V, Thm. 7)
    Invoked after Definition 3.1 to ensure the model is well-defined.
  • domain assumption Assumption 1: masks M_k^j are independent of times, counts, and spatial points; every component has positive observation probability
    Needed for the left-limit control in equation (13) and for the metric d_k to be meaningful.
  • domain assumption Assumption 2: X is almost surely not observed at a jump
    Avoids ambiguity in the left-limit loss; Remark 2.3 describes a weakening.
  • ad hoc to paper Assumption 3: existence of an integrable generalized derivative f such that F(t,O)=F(τ(t),O)+∫ f(s,O)ds
    This is the load-bearing regularity assumption that makes the evolution network representable; it is stronger than measurability and is not a generic property of all square-integrable càdlàg processes.
  • domain assumption Assumption 4: square integrability of the observed values
    Ensures the loss function is finite; used in the Monte Carlo proof.
  • domain assumption Assumption 5: integrability of n and sup_i J_i
    Used to show that ε-truncated subspaces exhaust the support as ε→0.
  • domain assumption Assumption 6: all observation points ξ_i^j are i.i.d. copies of ξ ~ μ_Ξ
    Used to relate averages over spatial points to a single random point; the proof also needs exchangeability of the j-indexed terms, which Assumption 6 alone does not fully provide.
  • domain assumption Assumption 7: X is independent of the observation framework
    Used throughout the variance decomposition and the metric control; can be weakened to conditional independence per Andersson et al. (2024).
  • standard math Uniform law of large numbers / epi-convergence lemma (Ledoux-Talagrand, Rubinstein-Shapiro) used in Lemma 4.6
    Basis for the Monte Carlo convergence of the empirical loss on compact parameter sets.

pith-pipeline@v1.3.0-alltime-deepseek · 37240 in / 15738 out tokens · 155375 ms · 2026-08-01T03:34:15.225253+00:00 · methodology

0 comments
read the original abstract

In this paper, we study the extension of Neural Jump ODEs to infinite-dimensional function spaces. In particular, the underlying process $X$ now takes values in $L^2(\Xi, \mathbb{R}^{d_X})$ instead of $\mathbb{R}^{d_X}$ and the Operator NJ-ODE approximates the optimal predictor of this process by producing a representative of the conditional expectation. The NJ-ODE model is a framework for online learning the optimal prediction of continuous-time stochastic processes, given discrete, possibly irregular and incomplete past observations. In a series of works, this model has been extended to deal with generic path-dependent processes, with observation noise and dependent observations, with long-term predictions, and with input-output systems. However, throughout all of these works, the underlying processes were restricted to be finite-dimensional. In particular, function-valued problems, like yield curve or volatility surface predictions, could only be handled through discretization, which inherently leads to a loss of information. In this work, we build on ideas from Neural Operator methods that allow us to extend the NJ-ODE framework to an infinite-dimensional output process. To prove convergence of the NJ-ODE to the optimal prediction process, we develop a new approximation strategy that also generalizes previous works in the finite-dimensional setting by considerably weakening the assumptions.

Figures

Figures reproduced from arXiv: 2607.23110 by Florian Krach, Josef Teichmann, Oliver L\"othgren.

Figure 1
Figure 1. Figure 1: (a) Example path of the Brownian cosine process, from [PITH_FULL_IMAGE:figures/full_fig_p023_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Evolution of the X(t, ξ) = Wt cos(ξ) path of minimal MSE, showcasing the O-NJ-ODE prediction and target conditional expectation. Although not shown here, performance is maintained until terminal time T. 28 [PITH_FULL_IMAGE:figures/full_fig_p028_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: O-NJ-ODE training diagnostics for the BrownianCosine dataset. [PITH_FULL_IMAGE:figures/full_fig_p029_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Leaky ReLU activations in the readout result in marginally slower and less stable training, with [PITH_FULL_IMAGE:figures/full_fig_p031_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Reducing the sample size only marginally degrades training and performance in absolute terms. [PITH_FULL_IMAGE:figures/full_fig_p033_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Despite the reduction in sample size to a quarter of the baseline, the decrease in prediction accuracy [PITH_FULL_IMAGE:figures/full_fig_p034_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Although small in absolute terms, increasing spatial data sparsity has the greatest impact on both [PITH_FULL_IMAGE:figures/full_fig_p035_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Prediction performance is noticeably weaker with the decreased access to spatial information, but [PITH_FULL_IMAGE:figures/full_fig_p036_9.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

129 extracted references · 2 canonical work pages

  1. [1]

    arXiv preprint arXiv:2510.02757 , year=

    Neural jump ODEs as generative models , author=. arXiv preprint arXiv:2510.02757 , year=

    arXiv

  2. [2]

    Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics , pages =

    Nonlinear functional regression: a functional RKHS approach , author =. Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics , pages =. 2010 , editor =

    2010

  3. [3]

    Proceedings of the Eleventh International Conference on Artificial Intelligence and Statistics , pages =

    Metric Learning for Kernel Regression , author =. Proceedings of the Eleventh International Conference on Artificial Intelligence and Statistics , pages =. 2007 , editor =

    2007

  4. [4]

    Gaussian Processes in Machine Learning

    Rasmussen, Carl Edward. Gaussian Processes in Machine Learning. Advanced Lectures on Machine Learning: ML Summer Schools 2003, Canberra, Australia, February 2 - 14, 2003, T \"u bingen, Germany, August 4 - 16, 2003, Revised Lectures. 2004

    2003

  5. [5]

    IEEE signal processing magazine , volume=

    Particle filtering , author=. IEEE signal processing magazine , volume=. 2003 , publisher=

    2003

  6. [6]

    Journal of basic Engineering , volume=

    A new approach to linear filtering and prediction problems [J] , author=. Journal of basic Engineering , volume=

  7. [7]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Structured inference networks for nonlinear state space models , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  8. [8]

    arXiv preprint arXiv:1511.07367 , year=

    Black box variational inference for state space models , author=. arXiv preprint arXiv:1511.07367 , year=

    Pith/arXiv arXiv

  9. [9]

    International conference on artificial intelligence and statistics , pages=

    Variational sequential monte carlo , author=. International conference on artificial intelligence and statistics , pages=. 2018 , organization=

    2018

  10. [10]

    arXiv preprint arXiv:1605.06432 , year=

    Deep variational bayes filters: Unsupervised learning of state space models from raw data , author=. arXiv preprint arXiv:1605.06432 , year=

    Pith/arXiv arXiv

  11. [11]

    Krishnan, Rahul G and Shalit, Uri and Sontag, David , journal=. Deep

  12. [12]

    IEEE Transactions on Signal Processing , volume=

    KalmanNet: Neural network aided Kalman filtering for partially known dynamics , author=. IEEE Transactions on Signal Processing , volume=. 2022 , publisher=

    2022

  13. [13]

    Tambe-Ndonfack , journal =

    Jakob Heiss and Florian Krach and Thorsten Schmidt and Félix B. Tambe-Ndonfack , journal =. Nonparametric filtering, estimation and classification using neural jump. doi:doi:10.1515/strm-2025-0001 , year =

    work page doi:10.1515/strm-2025-0001 2025

  14. [14]

    arXiv preprint arXiv:1607.06450 , year=

    Layer normalization , author=. arXiv preprint arXiv:1607.06450 , year=

    Pith/arXiv arXiv

  15. [15]

    Infinite Dimensional Analysis A Hitchhiker’s Guide , author=

  16. [16]

    arXiv preprint arXiv:2407.18808 , year=

    Learning Chaotic Systems and Long-Term Predictions with Neural Jump ODEs , author=. arXiv preprint arXiv:2407.18808 , year=

    Pith/arXiv arXiv

  17. [17]

    Journal of Machine Learning Research , volume=

    Neural operator: Learning maps between function spaces with applications to pdes , author=. Journal of Machine Learning Research , volume=

  18. [18]

    2021 , publisher=

    Foundations of modern probability , author=. 2021 , publisher=

    2021

  19. [19]

    and Rao, B

    Karandikar, Rajeeva L. and Rao, B. V. , year=. Introduction to Stochastic Calculus , ISBN=. doi:10.1007/978-981-10-8318-1 , publisher=

    work page doi:10.1007/978-981-10-8318-1

  20. [20]

    Math24 , year=

    Double Pendulum , author=. Math24 , year=

  21. [21]

    2021 , publisher=

    Saxena, Divya and Cao, Jiannong , journal=. 2021 , publisher=

    2021

  22. [22]

    2021 , organization=

    Kidger, Patrick and Foster, James and Li, Xuechen and Lyons, Terry J , booktitle=. 2021 , organization=

    2021

  23. [23]

    2021 , organization=

    Morrill, James and Salvi, Cristopher and Kidger, Patrick and Foster, James , booktitle=. 2021 , organization=

    2021

  24. [24]

    James Morrill and Patrick Kidger and Lingyi Yang and Terry Lyons , journal=

  25. [25]

    Extending Path-Dependent

    William Andersson and Jakob Heiss and Florian Krach and Josef Teichmann , journal=. Extending Path-Dependent

  26. [26]

    arXiv preprint arXiv:2206.14284 , year=

    Krach, Florian and N. arXiv preprint arXiv:2206.14284 , year=

    Pith/arXiv arXiv

  27. [27]

    A first course in dynamics: With a panorama of recent developments , DOI=

    Hasselblatt, Boris and Katok, Anatole , year=. A first course in dynamics: With a panorama of recent developments , DOI=

  28. [28]

    2009 , publisher=

    Fundamentals of stochastic filtering , author=. 2009 , publisher=

    2009

  29. [29]

    Algorithmic Finance , volume=

    Classification-based financial markets prediction using deep neural networks , author=. Algorithmic Finance , volume=. 2017 , publisher=

    2017

  30. [30]

    Applied Soft Computing , volume=

    Using deep learning for price prediction by exploiting stationary limit order book features , author=. Applied Soft Computing , volume=. 2020 , publisher=

    2020

  31. [31]

    2017 IEEE 19th conference on business informatics (CBI) , volume=

    Forecasting stock prices from the limit order book using convolutional neural networks , author=. 2017 IEEE 19th conference on business informatics (CBI) , volume=. 2017 , organization=

    2017

  32. [32]

    IEEE Transactions on Emerging Topics in Computational Intelligence , volume=

    Temporal bag-of-features learning for predicting mid price movements using high frequency limit order book data , author=. IEEE Transactions on Emerging Topics in Computational Intelligence , volume=. 2018 , publisher=

    2018

  33. [33]

    2017 IEEE Symposium Series on Computational Intelligence (SSCI) , pages=

    Tensor representation in high-frequency financial data for price change prediction , author=. 2017 IEEE Symposium Series on Computational Intelligence (SSCI) , pages=. 2017 , organization=

    2017

  34. [34]

    Journal of Forecasting , volume=

    Benchmark dataset for mid-price forecasting of limit order book data with machine learning methods , author=. Journal of Forecasting , volume=. 2018 , publisher=

    2018

  35. [35]

    IEEE Transactions on Signal Processing , volume=

    Deeplob: Deep convolutional neural networks for limit order books , author=. IEEE Transactions on Signal Processing , volume=. 2019 , publisher=

    2019

  36. [36]

    2021 , url=

    High Frequency Crypto Limit Order Book Data , author=. 2021 , url=

    2021

  37. [37]

    Bounded Variation and Around , year=

    Bounded variation and around , author=. Bounded Variation and Around , year=

  38. [38]

    Advances in Psychology , volume=

    Serial order: A parallel distributed processing approach , author=. Advances in Psychology , volume=. 1997 , publisher=

    1997

  39. [39]

    1985 , institution=

    Learning internal representations by error propagation , author=. 1985 , institution=

    1985

  40. [40]

    Advances in Neural Information Processing Systems , volume=

    Brits: Bidirectional recurrent imputation for time series , author=. Advances in Neural Information Processing Systems , volume=

  41. [41]

    Scientific Reports , volume=

    Recurrent neural networks for multivariate time series with missing values , author=. Scientific Reports , volume=

  42. [42]

    arXiv , year=

    The iisignature library: Efficient calculation of iterated-integral signatures and log signatures , author=. arXiv , year=

  43. [43]

    Probability and Measure , year =

    Billingsley, Patrick , edition =. Probability and Measure , year =

  44. [44]

    Multivariate statistics: A vector space approach , year =

    Eaton, Morris L , publisher =. Multivariate statistics: A vector space approach , year =

  45. [45]

    Journal of Mathematical Analysis and Applications , volume =

    On the characterization of compactness in the space of functions of bounded variation in the sense of. Journal of Mathematical Analysis and Applications , volume =. 2020 , author =

    2020

  46. [46]

    2019 , journal=

    Deep Signature Transforms , author=. 2019 , journal=

    2019

  47. [47]

    2020 , journal=

    Embedding and learning with signatures , author=. 2020 , journal=

    2020

  48. [48]

    Kiraly and Harald Oberhauser , title =

    Franz J. Kiraly and Harald Oberhauser , title =. Journal of Machine Learning Research , year =

  49. [49]

    and Caruana, Michael J

    Lyons, Terry J. and Caruana, Michael J. and Lévy, Thierry , year =. Differential. doi:10.1007/978-3-540-71285-5 , file =

    doi:10.1007/978-3-540-71285-5

  50. [50]

    Uniqueness for the signature of a path of bounded variation and the reduced path group , volume=

    Hambly, Ben and Lyons, Terry , year=. Uniqueness for the signature of a path of bounded variation and the reduced path group , volume=. Annals of Mathematics , publisher=. doi:10.4007/annals.2010.171.109 , number=

    doi:10.4007/annals.2010.171.109 2010

  51. [51]

    2016 , journal=

    Learning from the past, predicting the statistics for the future, learning an evolving system , author=. 2016 , journal=

    2016

  52. [52]

    2016 , journal=

    A Primer on the Signature Method in Machine Learning , author=. 2016 , journal=

    2016

  53. [53]

    2018 , eprint=

    Derivatives pricing using signature payoffs , author=. 2018 , eprint=

    2018

  54. [54]

    2016 , eprint=

    Kernels for sequentially ordered data , author=. 2016 , eprint=

    2016

  55. [55]

    Finance and Stochastics , volume=

    Moment explosions in stochastic volatility models , author=. Finance and Stochastics , volume=. 2007 , publisher=

    2007

  56. [56]

    Mathematical Finance: An International Journal of Mathematics, Statistics and Financial Economics , volume=

    Moment explosions and long-term behavior of affine stochastic volatility models , author=. Mathematical Finance: An International Journal of Mathematics, Statistics and Financial Economics , volume=. 2011 , publisher=

    2011

  57. [57]

    BOUNDS AND ESTIMATORS OF A BASIC CONSTANT IN EXTREME VALUE THEORY OF

    Qi-Man Shao , journal =. BOUNDS AND ESTIMATORS OF A BASIC CONSTANT IN EXTREME VALUE THEORY OF

  58. [58]

    Statistics & Probability Letters , volume =

    A note on upper estimates for. Statistics & Probability Letters , volume =. 2008 , author =

    2008

  59. [59]

    Stochastics , volume =

    Konstantin Borovkov and Yuliya Mishura and Alexander Novikov and Mikhail Zhitlukhin , title =. Stochastics , volume =. 2017 , publisher =

    2017

  60. [60]

    , address =

    Mohammed, S.E.A. , address =. Stochastic functional differential equations , year =

  61. [61]

    Bernoulli , number =

    Ilkka Norros and Esko Valkeila and Jorma Virtamo , title =. Bernoulli , number =

  62. [62]

    Prediction Law of fractional

    Tommi Sottinen and Lauri Viitasaari , year=. Prediction Law of fractional

  63. [63]

    Taraldsen, Gunnar , journal=

  64. [64]

    Springer , Title =

    Cohen, Samuel N and Elliott, Robert James , Date-Added =. Springer , Title =

  65. [65]

    Goldberger and Luis A

    Ary L. Goldberger and Luis A. N. Amaral and Leon Glass and Jeffrey M. Hausdorff and Plamen Ch. Ivanov and Roger G. Mark and Joseph E. Mietus and George B. Moody and Chung-Kang Peng and H. Eugene Stanley , Date-Added =. Circulation , Title =

  66. [66]

    Andersen and Tim Bollerslev , Date-Added =

    Torben G. Andersen and Tim Bollerslev , Date-Added =. Journal of Empirical Finance , Title =. 1997 , Bdsk-Url-1 =

    1997

  67. [67]

    arXiv:1710.02196 , Title =

    Soheil Feizi and Hamid Javadi and Jesse Zhang and David Tse , Date-Added =. arXiv:1710.02196 , Title =

    Pith/arXiv arXiv

  68. [68]

    Advances in Neural Information Processing Systems , Title =

    Lecchini-Visintini, Andrea and Lygeros, John and Maciejowski, Jan , Date-Added =. Advances in Neural Information Processing Systems , Title =

  69. [69]

    Journal of Optimization Theory and applications , Title =

    Locatelli, Marco , Date-Added =. Journal of Optimization Theory and applications , Title =

  70. [70]

    International Conference on Learning Representations , year=

    Neural Jump Ordinary Differential Equations: Consistent Continuous-Time Prediction and Filtering , author=. International Conference on Learning Representations , year=

  71. [71]

    Monte Carlo Methods and Applications , Title =

    Jean-Fran. Monte Carlo Methods and Applications , Title =

  72. [72]

    Efficient Simulation of the Heston Stochastic Volatility Model , Year =

    Andersen, Leif , Journal =. Efficient Simulation of the Heston Stochastic Volatility Model , Year =

  73. [73]

    Islam and C

    K. Islam and C. Shelton , Date-Modified =. Neural Stochastic Differential Equations with Bayesian Jumps for Marked Temporal Point Process , Year =

  74. [74]

    Learning Long-Term Dependencies in Irregularly-Sampled Time Series , Year =

  75. [75]

    A simplified analytical approach for pricing discretely-sampled variance swaps with stochastic volatility , Year =

    Sanae Rujivan and Song-Ping Zhu , Journal =. A simplified analytical approach for pricing discretely-sampled variance swaps with stochastic volatility , Year =

  76. [76]

    Menne, M. J. and Williams, Jr., C. N. and Vose, R. S. , Title =

  77. [77]

    Monte Carlo Methods and Applications , number =

    Bernard Lapeyre and Jérôme Lelong , title =. Monte Carlo Methods and Applications , number =. 2021 , pages =

    2021

  78. [78]

    Wiley , Title =

    Rubinstein, Reuven Y and Shapiro, Alexander , Date-Modified =. Wiley , Title =

  79. [79]

    Probability in Banach spaces , Volume =

    Ledoux, Michel and Talagrand, Michel , Journal =. Probability in Banach spaces , Volume =

  80. [80]

    Advances in Neural Information Processing Systems , volume=

    Neural controlled differential equations for irregular time series , author=. Advances in Neural Information Processing Systems , volume=

Showing first 80 references.